Function adam 

pub fn adam(
    f: impl Fn(&[f64]) -> f64,
    grad: impl Fn(&[f64]) -> Vec<f64>,
    x0: Vec<f64>,
    lr: f64,
    beta1: f64,
    beta2: f64,
    eps: f64,
    max_iter: u32,
) -> OptResult

Adam adaptive-moment optimizer.

Uses the standard Adam update rule with bias correction. Stops when the L2 norm of the gradient falls below the default tolerance 1e-7, or after max_iter steps (no separate tol parameter because Adam is typically used for a fixed budget).